Vision-based motion estimation for mobile robot navigation in harsh environment

Abstract

Visual Odometry (VO) is a process of estimating the pose of a moving agent relying solely on analysis of the image streams from one or multiple onboard cameras. It finds diverse applications in Robotics Navigation for both outdoor and Indoor settings, Environmental 3D Reconstruction, Autonomous Vehicles, Augmented Reality (AR), Virtual Reality (VR), and Camera Stabilization. It facilitates safe and efficient navigation, exploration, and interaction with the environment. In outdoor scenarios, stereo visual odometry enables navigating through varied terrains making informed decisions for obstacle avoidance and path planning. This research work focuses on developing a robust stereo visual odometry algorithm for mobile robot navigation in outdoor environments, which can be further complimented by external sensors. The research's primary objective is to create a stereo vision algorithm robustly estimating robot motion based on stereo camera data. The algorithm incorporates state-of-the-art feature detection, tracking, and filtering methods, enhancing the robot's environmental perception and navigation efficiency. By investigating these techniques and integrating them into the stereo visual odometry algorithm, the research aims to improve accuracy and robustness outdoor environments. Extensive real-world dataset evaluations confirm the algorithm's effectiveness in providing incremental online estimation of the robot's position and orientation, showcasing its potential for safer and more reliable navigation in various real-world applications. However, we also identified certain shortcomings during these tests and have diligently sought solutions for them. We rigorously evaluated the algorithm's effectiveness through individual assessments, aiming to uncover its strengths and limitations. These assessments shed light on its performance across a spectrum of real-world scenarios, providing valuable insights for further refinement and optimization. By substantiating its performance through real-world datasets and self-conducted experiments, this research offers comprehensive evidence of its potential. It significantly contributes to the field of stereo visual odometry by introducing a scalable and efficient algorithm. This innovation stands to advance robotics, autonomous vehicles, and augmented reality systems, ultimately improving our ability to seamlessly explore and interact with outdoor environments.